I’ve been experimenting with creating CustomGPTs using the ChatGPT Plus account at my internship and managed to build a really effective one. This inspired me to upgrade my personal account to ChatGPT Plus so I could continue creating CustomGPTs without cluttering my internship account or losing access to my work once the internship ends.
However, I’m facing an issue. Even when I use the exact same instructions from the CustomGPT I created on my internship account, the CustomGPT on my personal account performs noticeably worse. It often fails to follow even the basic instructions, and I’ve tested this across multiple chats. The pattern is consistent—responses on my internship account are consistently better than on my personal account.
This is really frustrating because I want to keep my work separate and ensure I can continue using my CustomGPTs after my internship. While I know I could share the link to the CustomGPT from my internship account and use it on my personal account, that’s not a viable solution. The bot could be deleted once my internship ends, and, realistically, the CustomGPT on my personal account should perform the same way if it’s following identical instructions.
Does anyone have any idea why this is happening or how to resolve it? Any help would be greatly appreciated!
You are so right!
I have given very detailed instructions to my customGPT but it behaves differently every time and on different accounts.
For example: I have asked GPT to get multiple inputs from the user and give output in stages. It doesn’t follow my instructions, sometimes skips steps or sometimes gives entire output in one go.
It’s frustrating
We are seeing same issue. Started sometime in last few weeks. Previously plus or free accounts provided similar responses (for practical purposes identical). Now getting very different responses depending on paid v free (though going to try someone else’s paid account). In short we have a customGPT that is instructed to ONLY source a PDF document for information. It is explicitly instructed not to search its “internal knowledge,” the web, etc. It works perfectly on my plus account but is giving out information that is both (1) inaccurate and (2) not in the PDF document to the free account. Despite updating its instructions, still the difference. Seems to be since an update in the last few weeks. Any help is greatly appreciated:
I wonder if it has something to do with the “READ MOD Guide (June 2022).pdf” in the incorrect response. I did not see that step (“thinking”) in the plus/correct response.
